To establish the optimum visible-near infrared (NIR) spectroscopy analysis model of total bilirubin (TBIL),direct bilirubin (DBIL) and indirect bilirubin (IBIL) in human serum,the modeling region is opt imized by combining transmitted spectrum technology with interval partial least squares (iPLS) or synergy interval partia l least squares (SiPLS) to choose characteristic bands of serum spectrum.Quantitative prediction models of the relationship betw een spectrum and three serum proteins are established accordingly.Root-mean-square error (RMSE) prediction was used as an evaluation standard of models.Results demonstrate that the SiPLS prediction model is superior to the iPLS prediction model.In the SiPLS prediction model,the optimum modeling wavelength ranges of TBIL,DBIL and IBIL in serum are 400~536nm ,1366~1502nm,2324~2460nm and 400~502nm,608~710nm,1644~1746nm,as well as 1644~1746nm,400~502nm,1746~1848nm,respectively.RMSEs of these three proteins are 0.5989μmol/L,0.2072μmol/L and 0.3862μmol/L,respectively.Waveband optimization is of important significance to improve prediction accuracy.The SiPLS-based quantitative analy sis model of TBIL,DBIL and IBIL in serum has high prediction accuracy and overcomes shortcomings of the iPLS-based model.Fu rthermore,the optimized characteristic spectrum region can provide references for designing small spectrophotometer s for special purposes.